Can nature’s own design process help the next big AI leap?

ITU professor Sebastian Risi is the co-author of a new open-access resource on an emerging field that could shape the future of artificial intelligence.

Sebastian Risi Collaborations Research artificial intelligence

Written 19 November, 2025 07:35 by Theis Duelund Jensen

Along with colleagues and industry partners, Professor Sebastian Risi of the IT University of Copenhagen has published a new textbook, Neuroevolution: Harnessing Creativity in AI Agent Design, on neuroevolution, a growing field in artificial intelligence that combines evolutionary principles with neural networks. The book has been more than four years in the making and is now available in an open-access HTML format. A physical edition will be published by MIT Press next year.

Neuroevolution takes inspiration from nature’s own design process – evolution – to create and optimise artificial neural networks. Instead of relying on traditional supervised learning, neuroevolution uses evolutionary algorithms to improve AI systems over generations, enabling them to discover creative solutions to complex problems.

“We’ve been working on this book for over four years, and it’s exciting to finally share it with the world,” says Sebastian Risi. “The basic idea is to take inspiration from how intelligence emerged in nature. Evolution invented nervous systems and optimised them over millions of years. We replicate that process in simulations to evolve neural networks. It’s a very creative force that often finds solutions we wouldn’t have thought of ourselves.”

Neuroevolution has been applied across a wide range of domains—from evolving neural networks that learn to play video games such as Doom, Unreal, and Super Mario, to enabling robots to adapt to unforeseen situations in the real world. Beyond control, its versatility and creative potential make it suitable for many other tasks, including designing soft-robot morphologies, generating video-game levels, and even supporting experiments that help us explore the origins of intelligence in nature.

The book covers the foundations of evolutionary algorithms and neural networks, explores their integration, and highlights applications ranging from robotics and generative AI to reinforcement learning and architecture search. It also addresses future challenges and opportunities for the field.

“Current AI systems, like large language models, are incredibly powerful but they also reveal limitations,” says Sebastian Risi. “We believe that combining these systems with evolutionary principles could be the next big leap toward more general and embodied intelligence. Evolution is the only process we know that has created real intelligence, and it remains a uniquely creative approach.”

The textbook features a foreword by renowned complexity scientist Melanie Mitchell, author of Artificial Intelligence: A Guide for Thinking Humans. Co-authors include leading figures from industry and academia, such as Yujin Tang and David Ha of Sakana AI and Risto Miikkulainen of Cognizant, reflecting the growing interest in neuroevolution across both research and commercial sectors.

Beyond serving as a teaching resource, the authors hope the book will help build a larger community around neuroevolution. “AI research today is dominated by a monoculture of large language models,” says Sebastian Risi. “We want to inspire the next generation to explore alternative approaches. Diversity of ideas is crucial for progress.”